Antibody Structure
A standard antibody (Ab) is a tetrameric structure consisting of two identical immunoglobulin (Ig) heavy chains and two identical light chains. The heavy and light chains of an Ab consist of different domains. Each light chain has one variable domain (VL) and one constant domain (CL), while each heavy chain has one variable domain (VH) and three or four constant domains (CH) (Alzari et al., 1988). Each domain, consisting of ˜110 amino acid residues, is folded into a characteristic β-sandwich structure formed from two β-sheets packed against each other, the immunoglobulin fold. The VH and VL domains each have three complementarity  determining regions (CDR1-3) that are loops, or turns, connecting β-strands at one end of the domains (FIG. 1: A, C). The variable regions of both the light and heavy chains generally contribute to antigen specificity, although the contribution of the individual chains to specificity is not always equal. Antibody molecules have evolved to bind to a large number of molecules by using six randomized loops (CDRs). However, the size of the antibodies and the complexity of six loops represents a major design hurdle if the end result is to be a relatively small peptide ligand.
Antibody Substructures
Functional substructures of Abs can be prepared by proteolysis and by recombinant methods. They include the Fab fragment, which comprises the VH-CH1 domains of the heavy chain and the VL-CL1 domains of the light chain joined by a single interchain disulfide bond, and the Fv fragment, which comprises only the VH and VL domains. In some cases, a single VH domain retains significant affinity (Ward et al., 1989). It has also been shown that a certain monomeric κ light chain will specifically bind to its cognate antigen. (L. Masat et al., 1994). Separated light or heavy chains have sometimes been found to retain some antigen-binding activity (Ward et al., 1989). These antibody fragments are not suitable for structural analysis using NMR spectroscopy due to their size, low solubility or low conformational stability.
Another functional substructure is a single chain Fv (scFv), comprised of the variable regions of the immunoglobulin heavy and light chain, covalently connected by a peptide linker (S-z Hu et al., 1996). These small (M, 25,000) proteins generally retain specificity and affinity for antigen in a single polypeptide and can provide a convenient building block for larger, antigen-specific molecules. Several groups have reported biodistribution studies in xenografted athymic mice using scFv reactive against a variety of tumor antigens, in which specific tumor localization has been observed. However, the short persistence of scFvs in the circulation limits the  exposure of tumor cells to the scFvs, placing limits on the level of uptake. As a result, tumor uptake by scFvs in animal studies has generally been only 1-5% ID/g as opposed to intact antibodies that can localize in tumors ad 30-40% ID/g and have reached levels as high as 60-70% ID/g.
A small protein scaffold called a “minibody” was designed using a part of the Ig VH domain as the template (Pessi et al., 1993). Minibodies with high affinity (dissociation constant (Kd)˜10−7 M) to interleukin-6 were identified by randomizing loops corresponding to CDR1 and CDR2 of VH and then selecting mutants using the phage display method (Martin et al., 1994). These experiments demonstrated that the essence of the Ab function could be transferred to a smaller system. However, the minibody had inherited the limited solubility of the VH domain (Bianchi et al., 1994).
It has been reported that camels (Camelus dromedarius) often lack variable light chain domains when IgG-like material from their serum is analyzed, suggesting that sufficient antibody specificity and affinity can be derived form VH domains (three CDR loops) alone. Davies and Riechmann recently demonstrated that “camelized” VH domains with high affinity (Kd˜10−7 M) and high specificity can be generated by randomizing only the CDR3. To improve the solubility and suppress nonspecific binding, three mutations were introduced to the framework region (Davies & Riechmann, 1995). It has not been definitively shown, however, that camelization can be used, in general, to improve the solubility and stability of VHs.
An alternative to the “minibody” is the “diabody.” Diabodies are small bivalent and bispecific antibody fragments, i.e., they have two antigen-binding sites. The fragments comprise a heavy-chain variable domain (VH) connected to a light-chain variable domain (VL) on the same polypeptide chain (VH-VL). Diabodies are similar in size to an Fab fragment. By using a linker that is too short to allow pairing between the two domains on the same chain, the domains are forced to pair with the complementary domains of another chain and create two antigen-binding  sites. These dimeric antibody fragments, or “diabodies,” are bivalent and bispecific. P. Holliger et al., PNAS 90:6444-6448 (1993).
Since the development of the monoclonal antibody technology, a large number of 3D structures of Ab fragments in the complexed and/or free states have been solved by X-ray crystallography (Webster et al., 1994; Wilson & Stanfield, 1994). Analysis of Ab structures has revealed that five out of the six CDRs have limited numbers of peptide backbone conformations, thereby permitting one to predict the backbone conformation of CDRs using the so-called canonical structures (Lesk & Tramontano, 1992; Rees et al., 1994). The analysis also has revealed that the CDR3 of the VH domain (VH-CDR3) usually has the largest contact surface and that its conformation is too diverse for canonical structures to be defined; VH-CDR3 is also known to have a large variation in length (Wu et al., 1993). Therefore, the structures of crucial regions of the Ab-antigen interface still need to be experimentally determined.
Comparison of crystal structures between the free and complexed states has revealed several types of conformational rearrangements. They include side-chain rearrangements, segmental movements, large rearrangements of VH-CDR3 and changes in the relative position of the VH and VL domains (Wilson & Stanfield, 1993). In the free state, CDRs, in particular those which undergo large conformational changes upon binding, are expected to be flexible. Since X-ray crystallography is not suited for characterizing flexible parts of molecules, structural studies in the solution state have not been possible to provide dynamic pictures of the conformation of antigen-binding sites.
Mimicking the Antibody-binding Site
CDR peptides and organic CDR mimetics have been made (Dougall et al., 1994). CDR peptides are short, typically cyclic, peptides which correspond to the amino acid sequences of CDR loops of antibodies. CDR loops are responsible for  antibody-antigen interactions. Organic CDR mimetics are peptides corresponding to CDR loops which are attached to a scaffold, e.g., a small organic compound.
CDR peptides and organic CDR mimetics have been shown to retain some binding affinity (Smyth & von Itzstein, 1994). However, as expected, they are too small and too flexible to maintain full affinity and specificity. Mouse CDRs have been grafted onto the human Ig framework without the loss of affinity (Jones et al., 1986; Riechmann et al., 1988), though this “humanization” does not solve the above-mentioned problems specific to solution studies.
Mimicking Natural Selection Processes of Abs
In the immune system, specific Abs are selected and amplified from a large library (affinity maturation). The processes can be reproduced in vitro using combinatorial library technologies. The successful display of Ab fragments on the surface of bacteriophage has made it possible to generate and screen a vast number of CDR mutations (McCafferty et al., 1990; Barbas et al., 1991; Winter et al., 1994). An increasing number of Fabs and Fvs (and their derivatives) is produced by this technique, providing a rich source for structural studies. The combinatorial technique can be combined with Ab mimics.
A number of protein domains that could potentially serve as protein scaffolds have been expressed as fusions with phage capsid proteins. Review in Clackson & Wells, Trends Biotechnol. 12:173-184 (1994). Indeed, several of these protein domains have already been used as scaffolds for displaying random peptide sequences, including bovine pancreatic trypsin inhibitor (Roberts et al., PNAS 89:2429-2433 (1992)), human growth hormone (Lowman et al., Biochemistry 30:10832-10838 (1991)), Venturini et al., Protein Peptide Letters 1:70-75 (1994)), and the IgG binding domain of Streptococcus (O'Neil et al., Techniques in Protein Chemistry V (Crabb, L, ed.) pp. 517-524, Academic Press, San Diego (1994)). These scaffolds have displayed a single randomized loop or region. 
Researchers have used the small 74 amino acid α-amylase inhibitor Tendamistat as a presentation scaffold on the filamentous phage M13 (McConnell and Hoess, 1995). Tendamistat is a β-sheet protein from Streptomyces tendae. It has a number of features that make it an attractive scaffold for peptides, including its small size, stability, and the availability of high resolution NMR and X-ray structural data. Tendamistat's overall topology is similar to that of an immunoglobulin domain, with two β-sheets connected by a series of loops. In contrast to immunoglobulin domains, the β-sheets of Tendamistat are held together with two rather than one disulfide bond, accounting for the considerable stability of the protein. By analogy with the CDR loops found in immunoglobulins, the loops the Tendamistat may serve a similar function and can be easily randomized by in vitro mutagenesis.
Tendamistat, however, is derived from Streptomyces tendae. Thus, while Tendamistat may be antigenic in humans, its small size may, reduce or inhibit its antigenicity. Also, Tendamistat's stability is uncertain. Further, the stability that is reported for Tendamistat is attributed to the presence of two disulfide bonds. Disulfide bonds, however, are a significant disadvantage to such molecules in that they can be broken under reducing conditions and must be properly formed in order to have a useful protein structure. Further, the size of the loops in Tendamistat are relatively small, thus limiting the size of the inserts that can be accommodated in the scaffold. Moreover, it is well known that forming correct disulfide bonds in newly synthesized peptides is not straightforward. When a protein is expressed in the cytoplasmic space of E. coli, the most common host bacterium for protein overexpression, disulfide bonds are usually not formed, potentially making it difficult to prepare large quantities of engineered molecules.
Thus, there is an on-going need for small, single-chain artificial antibodies for a variety of therapeutic, diagnostic and catalytic applications. 